Click rate prediction model training method and advertisement card display method

Abstract

The embodiment of the invention relates to a training method of a click rate prediction model and a display method of an advertisement card. The training of the click rate prediction model comprises the following steps: acquiring sample card information, user characteristics of a first user, a sample click rate and a plurality of style conflict maps corresponding to a plurality of style layers; performing hierarchical information extraction on the sample card information according to the plurality of style layers to obtain a plurality of sample style sets corresponding to the plurality of style layers; if the plurality of sample style sets pass the detection of the corresponding style conflict maps, inputting the user characteristics of the first user and the plurality of sample style sets into an initial model, outputting a predicted click rate of the first user by the initial model, and adjusting model parameters of the initial model according to the predicted click rate and the sample click rate, and when a preset convergence condition is met, taking the initial model as a click rate prediction model. Therefore, the matching effect between different styles in the advertisement card is ensured, and the advertisement card conforming to the user characteristics is displayed in real time.

Description

Technical Field

[0001] This disclosure relates to the field of computer technology, and in particular to a method for training a click-through rate prediction model and a method for displaying advertising cards. Background Technology

[0002] In the field of recommendation information, the display style of ad cards on the page directly affects the click-through rate (CTR) of users to the ad cards. However, there is a problem of mutual exclusion between different style parameter values.

[0003] In related technologies, due to the large number of style parameter values, in order to achieve real-time display of ad cards while satisfying the mutual exclusion of style parameter values, only some style parameters can be set to fixed style parameter values. Therefore, it is difficult to generate ad cards that adapt to user characteristics in real time while satisfying the mutual exclusion of styles. Summary of the Invention

[0004] To address the aforementioned technical issues, this disclosure provides a method for training a click-through rate prediction model and a method for displaying advertising cards.

[0005] This disclosure provides a method for training a click-through rate (CTR) prediction model, the method comprising: The sample card information of the sample ad card, the user characteristics of the first user, the sample click-through rate of the first user on the sample ad card, and multiple style conflict graphs corresponding to multiple style layers are obtained; wherein, the sample card information includes multiple style information, the multiple style layers are functional layers of the style information, and the style conflict graph is used to record the mutual exclusion relationship between style information. Based on the multiple style layers, the sample card information is extracted hierarchically to obtain multiple sample style sets corresponding to the multiple style layers; wherein, each sample style set includes style information corresponding to one style layer; If all of the multiple sample style sets pass the detection of the corresponding style conflict graph, then the user features of the first user and the multiple sample style sets are input into the initial model, and the initial model outputs the predicted click-through rate of the first user. The model parameters of the initial model are adjusted according to the predicted click-through rate of the first user and the sample click-through rate until a preset convergence condition is reached, and then the initial model is used as the click-through rate prediction model.

[0006] This disclosure also provides a method for displaying advertising cards, applied to an electronic device deployed with a click-through rate prediction model, the method comprising: In response to an ad display request sent by the client, obtain the user characteristics of the target user and the information of the ad card to be analyzed; Based on the layered information extraction of the card information to be analyzed according to multiple style layers, multiple sets of styles to be analyzed corresponding to the multiple style layers are obtained; The user characteristics of the target user and the multiple sets of styles to be analyzed are input into the click-through rate prediction model, and the click-through rate prediction model outputs the predicted click-through rate of the target user; the click-through rate prediction model is trained by the click-through rate prediction model training method. The target card information in the card information to be analyzed is determined based on the predicted click-through rate. The target card information is sent to the client so that the target advertising card is displayed on the client's page based on the target card information.

[0007] This disclosure also provides a training apparatus for a click-through rate prediction model, the apparatus comprising: The first acquisition module is used to acquire sample card information of sample advertising cards, user characteristics of the first user, sample click-through rate of the first user on the sample advertising cards, and multiple style conflict graphs corresponding to multiple style layers; wherein, the sample card information includes multiple style information, the multiple style layers are functional layers of the style information, and the style conflict graph is used to record the mutual exclusion relationship between style information. The first extraction module is used to extract information from the sample card information in layers according to the multiple style layers, so as to obtain multiple sample style sets corresponding to the multiple style layers; wherein, each sample style set includes style information corresponding to one style layer; The training module is used to input the user features of the first user and the multiple sample style sets into an initial model if all of the multiple sample style sets pass the detection of the corresponding style conflict map. The initial model outputs the predicted click-through rate of the first user. The model parameters of the initial model are adjusted according to the predicted click-through rate of the first user and the sample click-through rate until a preset convergence condition is reached. The initial model is then used as the click-through rate prediction model.

[0008] This disclosure also provides an advertising card display device, disposed on an electronic device equipped with a click-through rate prediction model, the device comprising: The second acquisition module is used to respond to the ad display request sent by the client and acquire the user characteristics of the target user and the information of the ad card to be analyzed. The second extraction module is used to extract information from the card information to be analyzed in layers according to multiple style layers, so as to obtain multiple sets of styles to be analyzed corresponding to the multiple style layers. The prediction module is used to input the user characteristics of the target user and the multiple sets of styles to be analyzed into the click-through rate prediction model, and the click-through rate prediction model outputs the predicted click-through rate of the target user; the click-through rate prediction model is trained by the click-through rate prediction model training method of any one of claims 1-9; The determination module is used to determine the target card information in the card information to be analyzed based on the predicted click-through rate; The display module is used to send the target card information to the client so that the target advertising card is displayed on the client's page according to the target card information.

[0009] This disclosure also provides an electronic device, the electronic device comprising: a processor; a memory for storing executable instructions of the processor; the processor being configured to read the executable instructions from the memory and execute the instructions to implement a click-through rate prediction model training method or an ad card display method as provided in this disclosure.

[0010] This disclosure also provides a computer-readable storage medium storing a computer program for executing a training method for a click-through rate prediction model or a display method for an advertising card as provided in this disclosure.

[0011] Compared with the prior art, the technical solution provided in this disclosure has the following advantages: The training scheme for the click-through rate prediction model provided in this disclosure includes: acquiring sample card information of sample advertising cards, user characteristics of a first user, sample click-through rate of the first user on the sample advertising cards, and multiple style conflict graphs corresponding to multiple style layers; wherein, the sample card information includes multiple style information, the multiple style layers are functional hierarchies of style information, and the style conflict graphs are used to record the mutual exclusion relationships between style information; extracting information from the sample card information according to the multiple style layers to obtain multiple sample style sets corresponding to the multiple style layers; wherein, a sample style set includes style information corresponding to a style layer; if multiple sample style sets all pass the detection of the corresponding style conflict graphs, then the user characteristics of the first user and the multiple sample style sets are input into the initial model, the initial model outputs the predicted click-through rate of the first user, and the model parameters of the initial model are adjusted according to the predicted click-through rate of the first user and the sample click-through rate until a preset convergence condition is reached, and the initial model is used as the click-through rate prediction model.

[0012] Using the above technical solution, multiple style layers are obtained according to function, and the mutual exclusion relationship between style information within each style layer is recorded through a style conflict graph. Sample card information is extracted according to style layers to obtain the sample style set corresponding to each style layer. The sample style set of the corresponding style layer is detected by the style conflict graph, avoiding consideration of the mutual exclusion relationship between style information of different style layers, realizing hierarchical isolation of style conflicts, and reducing the number of style conflict combinations to be considered. The initial model is trained based on the multiple sample style sets that have passed the detection, the user characteristics of the first user, and the sample click-through rate, to obtain a model that can predict the click-through rate based on the sample style sets that have passed the detection and user characteristics. The trained model can predict a reasonable click-through rate for ad cards that do not have mutual exclusion style information within the style layer based on user characteristics, and predict a lower click-through rate for ad cards that have mutual exclusion style information. This avoids the ad cards displayed to users containing styles with poor matching effects, and achieves the goal of ensuring the matching effect between different styles in the ad cards while displaying ad cards that match the user characteristics in real time.

[0013] The training scheme for the click-through rate (CTR) prediction model provided in the disclosed embodiments is applied to an electronic device on which the CTR prediction model is deployed. The scheme includes: responding to an ad display request sent by a client; obtaining user characteristics of the target user and information about the ad card to be analyzed; performing layered information extraction on the information about the ad card to be analyzed according to multiple style layers to obtain multiple sets of styles to be analyzed corresponding to multiple style layers; inputting the user characteristics of the target user and the multiple sets of styles to be analyzed into the CTR prediction model, and having the CTR prediction model output the predicted CTR of the target user; training the CTR prediction model using a CTR prediction model training method; determining the target card information in the ad card information to be analyzed based on the predicted CTR; and sending the target card information to the client so that the target ad card is displayed on the client's page based on the target card information.

[0014] Using the above technical solution, the information of the card to be analyzed is extracted according to the style layer, resulting in multiple sets of styles to be analyzed corresponding to multiple style layers. These multiple sets of styles to be analyzed, along with the user characteristics of the target user, are input into a click-through rate (CTR) prediction model. This CTR prediction model can predict a reasonable CTR for ad cards that do not have mutually exclusive style information within the style layer, and a lower CTR for ad cards that do have mutually exclusive style information, based on the user characteristics. The target card information in the card information to be analyzed is determined based on the predicted CTR, and the target ad card is displayed on the client side based on this target card information. This avoids displaying ad cards with poor matching effects to the user, ensuring the matching effect between different styles in the ad card while displaying ad cards that match the user characteristics in real time. Attached Figure Description

[0015] The above and other features, advantages, and aspects of the embodiments of this disclosure will become more apparent from the accompanying drawings and the following detailed description. Throughout the drawings, the same or similar reference numerals denote the same or similar elements. It should be understood that the drawings are schematic, and the originals and elements are not necessarily drawn to scale.

[0016] Figure 1 A flowchart illustrating a training method for a click-through rate prediction model provided in this embodiment of the present disclosure; Figure 2 A flowchart illustrating another training method for a click-through rate prediction model provided in this embodiment of the disclosure; Figure 3 A schematic diagram illustrating the determination of a target style conflict map according to an embodiment of this disclosure; Figure 4 A flowchart illustrating a method for displaying an advertising card according to an embodiment of this disclosure; Figure 5 A flowchart illustrating another method for displaying an advertising card provided in this embodiment of the disclosure; Figure 6 A schematic diagram illustrating a method for displaying an advertising card according to an embodiment of this disclosure; Figure 7 A schematic diagram of the structure of a training device for a click-through rate prediction model provided in an embodiment of this disclosure; Figure 8 A schematic diagram of the structure of an advertising card display device provided in an embodiment of this disclosure; Figure 9 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this disclosure. Detailed Implementation

[0017] Embodiments of this disclosure will now be described in more detail with reference to the accompanying drawings. While some embodiments of this disclosure are shown in the drawings, it should be understood that this disclosure can be implemented in various forms and should not be construed as limited to the embodiments set forth herein. Rather, these embodiments are provided to provide a more thorough and complete understanding of this disclosure. It should be understood that the accompanying drawings and embodiments of this disclosure are for illustrative purposes only and are not intended to limit the scope of protection of this disclosure.

[0018] It should be understood that the steps described in the method embodiments of this disclosure may be performed in different orders and / or in parallel. Furthermore, the method embodiments may include additional steps and / or omit the steps shown. The scope of this disclosure is not limited in this respect.

[0019] The term "comprising" and its variations as used herein are open-ended inclusions, meaning "including but not limited to". The term "based on" means "at least partially based on". The term "one embodiment" means "at least one embodiment"; the term "another embodiment" means "at least one additional embodiment"; the term "some embodiments" means "at least some embodiments". Definitions of other terms will be given in the description below.

[0020] It should be noted that the concepts of "first" and "second" mentioned in this disclosure are used only to distinguish different devices, modules or units, and are not used to limit the order of functions performed by these devices, modules or units or their interdependencies.

[0021] It should be noted that the terms "a" and "a plurality of" used in this disclosure are illustrative rather than restrictive, and those skilled in the art should understand that, unless otherwise expressly indicated in the context, they should be understood as "one or more".

[0022] The names of messages or information exchanged between multiple devices in the embodiments of this disclosure are for illustrative purposes only and are not intended to limit the scope of such messages or information.

[0023] In the field of digital advertising, the display style of an ad directly impacts user click-through rates. Related technologies employ fixed templates or limited A / B testing for style optimization, but this method suffers from the following problems: First, there are coupling conflicts between multiple style parameter values. Ads consist of multiple visual or interactive styles such as color, font, and animation effects, and there are implicit aesthetic conflicts and functional incompatibilities between style values. For example, a high-saturation background with dark fonts reduces readability. Second, there is a combination explosion and a lack of personalization; traditional methods cannot efficiently traverse the massive combinations of parameter values ​​(theoretically exceeding 10). 6 This method (which only allows for localized style optimization through preset style values) is difficult to adapt to user profiles in real time.

[0024] The aforementioned issues make it impossible to determine ad cards that are compatible with user characteristics from all parameters while maintaining aesthetics and consistency.

[0025] To address the aforementioned technical problems, this disclosure provides a method for training a click-through rate prediction model, which will be described below with reference to specific embodiments.

[0026] Figure 1 This is a flowchart illustrating a training method for a click-through rate (CTR) prediction model provided in an embodiment of this disclosure. This method can be executed by a training device for the CTR prediction model, which can be implemented in software and / or hardware, and is generally integrated into an electronic device. Figure 1As shown, the method includes: Step 101: Obtain sample card information of sample advertising cards, user characteristics of the first user, sample click-through rate of the first user on sample advertising cards, and multiple style conflict graphs corresponding to multiple style layers; wherein, sample card information includes multiple style information, multiple style layers are functional layers of style information, and style conflict graphs are used to record the mutual exclusion relationships between style information.

[0027] The sample ad card can be an ad card used as a training sample. An ad card can be a visual virtual card on a page that recommends information to the user. This embodiment does not limit the type of ad displayed on the ad card; for example, the ad can be a product recommendation, a game ad, a brand ad, etc. The sample card information can be the card information corresponding to the sample ad card. Card information is the information on which the ad card is generated; one ad card can correspond to one piece of card information. Style information can be information that characterizes the style of the ad card; one piece of style information can include a style type and its corresponding style value, where the style value can be the specific value of the style type. One ad card can correspond to multiple pieces of style information.

[0028] The user characteristics of the first user can be user characteristics used as training samples, obtained with the authorization of the first user. This embodiment does not limit the user characteristics. For example, the user characteristics can be generated based on one or more of the user's age, device type, and historical behavior. The sample click-through rate can be the click-through rate used as training samples, and the click-through rate can characterize the probability that a user performs a triggered action on the ad card. The sample click-through rate can be determined by testing or set manually; this embodiment does not limit it.

[0029] A style layer can be a functional hierarchy of style information based on style type. A style layer can include at least one style type, and different style layers can include different style types. Style layers enable parameter decoupling of style types. A style type can be the type of style information; for example, if the style type is color, then the style information can be a specific color. Specifically, a style type can be a parameter type in an ad card, representing the smallest element (i.e., an atomic element) that affects the style of the ad card. This embodiment does not limit the style type; for example, the style type can include the type of user interface (UI) or the type of user experience (UE) interaction.

[0030] In one optional implementation, the style layer includes at least two of a base layer, a content layer, and an interaction layer. The base layer may include a layout grid style, which may include grid position and / or grid size. The content layer may include style types related to the ad card content, and may include one or more of image position, image size, text position, text size, font, and button shape. The interaction layer may include style types related to user interaction; for example, the interaction layer may include interactive animations and / or interactive logic. The user interaction includes, but is not limited to, one or more of clicking, swiping, and shaking.

[0031] Multiple style layers can exist in a hierarchical relationship, where the hierarchy represents the position of a style layer within that relationship. Taking a style layer consisting of a base layer, content layer, and interaction layer as an example, the base layer can be the top layer, the next level after the base layer can be the content layer, and the next level after the content layer can be the interaction layer. Optionally, this hierarchical division can be determined based on the loading order of data in different style layers during page loading.

[0032] A style conflict graph can be understood as a co-occurrence conflict graph of style information. The style conflict graph is a graph structure, including nodes and edges. Nodes represent style information within the corresponding style layer, and edges represent mutually exclusive relationships between two nodes, meaning that the two nodes cannot appear simultaneously. For example, style types in the content layer can include background color and font color. In the style conflict graph corresponding to the content layer, the node corresponding to a high-saturation background color and the node corresponding to a dark font color are connected by an edge. This style conflict graph can be used to structurally represent the non-linear mutually exclusive relationships between style information corresponding to the respective style layers.

[0033] A subgraph can be a partial graph structure of a style conflict graph. A subgraph can include two nodes and edges connecting each pair of nodes. This embodiment does not limit the construction method of the subgraph. For example, the subgraph can be generated based on pre-set style rule information, which records mutually exclusive style information in the form of a document. The subgraph can also be determined based on implicit conflict information mined from display anomaly logs. Taking the subgraph A{ij} as an example, A{ij}=1 can indicate that style information i and style information j are mutually exclusive.

[0034] In this embodiment, the training device for the click-through rate prediction model can traverse and combine multiple style information to obtain multiple sample card information. It also acquires the user characteristics of a first user, the sample click-through rate of the first user for each sample card information, and the style conflict graph corresponding to each style layer.

[0035] In some embodiments of this disclosure, the style conflict map includes a layer conflict map and / or a group conflict map; the layer conflict map corresponds one-to-one with the style layer, and the layer conflict map is used to record the general mutual exclusion relationship between style information within the corresponding style layer; the group conflict map corresponds to the information group in the style layer, an information group includes multiple style information in the corresponding style layer, and an information group corresponds to specific style information in the preceding style layer of the style layer, the preceding style layer being the style layer at the next higher level, and the group conflict map is used to record the mutual exclusion relationship between style information within the information group.

[0036] Among them, layer conflict maps can be used to record universal style information mutual exclusion relationships within a style layer. Group conflict maps can be used to record the mutual exclusion relationships of style information within a specific type of group in a style layer. An information group can include all or part of the style information within a style layer. A style group conflict map includes one layer conflict map and / or at least one group conflict map.

[0037] In this embodiment, in addition to the top-level style layer, each style layer has a corresponding preceding style layer. The style information in the preceding style layer corresponds to an information group in the current style layer, and this information group has a corresponding group conflict graph. This indicates that when the advertising card uses this style information, the detection of the style information will be triggered using the group conflict graph. For example, if the preceding style layer is a content layer and the current style layer is an interaction layer, and the content layer contains style A, style A can correspond to information group a in the interaction layer. Information group a includes style 1, style 2, and style 3, and there exists a group conflict graph that records the mutual exclusion relationships between style 1, style 2, and style 3.

[0038] Step 102: Extract information from the sample card information in layers according to multiple style layers to obtain multiple sample style sets corresponding to multiple style layers; wherein, a sample style set includes style information corresponding to a style layer.

[0039] A sample style set can include the style information corresponding to the sample card information in a style layer.

[0040] In this embodiment of the disclosure, for each style layer, the training device of the click-through rate prediction model can extract the style information belonging to that style layer from the style information included in the sample card information to obtain the sample style set corresponding to that style layer.

[0041] In some embodiments of this disclosure, information extraction is performed on sample card information according to multiple style layers to obtain multiple sample style sets corresponding to multiple style layers, including: for each style layer, among the multiple style information corresponding to the sample card information, style information whose style type matches the style layer is assigned to the sample style set corresponding to the style layer.

[0042] In this embodiment, each style layer has a corresponding style type. For each style layer, the click-through rate prediction model can obtain the style type corresponding to that style layer during training, and classify the style information belonging to the style type of that style layer into the sample style set corresponding to that style layer, thus obtaining the sample style set corresponding to each style layer. Therefore, the efficient determination of the sample style set is achieved through style type matching.

[0043] Step 103: If multiple sample style sets pass the detection of the corresponding style conflict map, the user features of the first user and multiple sample style sets are input into the initial model, and the initial model outputs the predicted click-through rate of the first user. The model parameters of the initial model are adjusted according to the predicted click-through rate of the first user and the sample click-through rate until the preset convergence condition is reached, and the initial model is used as the click-through rate prediction model.

[0044] The initial model can be an untrained neural network model. This embodiment does not limit the type of the initial model; for example, the initial model can be a graph neural network (GNN-Transformer) model. Optionally, the structural layers in the initial model may include an input layer, a graph coding (GNN) layer, a transformer coding layer, a feature fusion layer, an output layer, etc.

[0045] The predicted click-through rate (CTR) can be the click-through rate of a user's ad card predicted by the model. Model parameters can be numerical variables that can be adjusted through learning within the model. The preset convergence condition can be a pre-set condition for the model to complete training. This embodiment does not limit the preset convergence condition; for example, the preset convergence condition could be that the error is less than a corresponding threshold, the error fluctuation is less than a corresponding threshold, the training reaches a corresponding number of rounds, etc. The CTR prediction model can be a trained model used to predict the click-through rate of users on ad cards.

[0046] In this embodiment, the training device for the click-through rate (CTR) prediction model can detect sample style sets using a style conflict graph. If each sample style set passes the detection, multiple sample style sets are characterized using self-attention encoding or other methods to obtain corresponding features, which can be normalized vectors. Furthermore, the user features of the first user and the features corresponding to multiple sample style sets are input into the initial model, and the predicted CTR is calculated using the initial model. Further, the model parameters are adjusted using a backpropagation algorithm based on the error between the predicted CTR and the sample CTR until a preset convergence condition is met, completing the training of the initial model. This trained initial model is then used as the CTR prediction model. This error can be calculated using a weighted cross-entropy loss function.

[0047] Optionally, the style conflict graph can be used as prior knowledge for message passing in the model. Furthermore, to lower the predicted click-through rate (CTR) for styles with mutually exclusive relationships, the initial model can be trained using negative samples. Specifically, sample card information with at least one sample style set that fails the style conflict graph detection can be identified, and the CTR corresponding to that sample card information can be updated to a value less than a preset threshold. The initial model is then trained based on the user characteristics of the first user, multiple sample style sets, and the updated CTR.

[0048] In some embodiments of this disclosure, multiple sample style sets pass the detection of corresponding style conflict maps, including: for each sample style set, if it is determined from the corresponding style conflict map that the sample style set includes two style information with a mutual exclusion relationship, then the sample style set is determined to have failed the detection; otherwise, the sample style set is determined to have passed the detection.

[0049] In this embodiment, for each sample style set, the training device of the click-through rate prediction model can extract style information from that sample style set and match the extracted style information with the style conflict graph corresponding to that style layer. If it is determined that there are two style information pieces in the style conflict graph that are connected as two nodes by an edge, then the sample style set is determined to have failed the detection; otherwise, the sample style set is considered to have passed the detection. Thus, efficient filtering of sample card information is achieved by using the nodes and edges recorded in the style conflict graph.

[0050] Figure 2 A flowchart illustrating another training method for a click-through rate prediction model provided in this disclosure embodiment is shown below. Figure 2 As shown, in some embodiments of this disclosure, determining the sample style set based on the corresponding style conflict map includes two style information pieces with a mutually exclusive relationship, including: Step 201: Extract pairwise style information from the sample style set to form multiple candidate style pairs.

[0051] Among them, the candidate style pair can be a set that includes any two style information from the sample style set.

[0052] In this embodiment, the training device for the click-through rate prediction model can extract pairwise style information from the sample style set and pair the pairwise style information to obtain multiple candidate style pairs.

[0053] Step 202: For each candidate style pair, determine the target style conflict graph in the multiple style conflict graphs that uses the two style information in the candidate style pair as nodes, and determine whether the two style information in the target style conflict graph are connected by an edge.

[0054] Among them, the target style conflict graph can be a style conflict graph in which the two style information in the candidate style pair are respectively used as two nodes.

[0055] In this embodiment, for each candidate style pair, the training device of the click-through rate prediction model can determine whether there exists a style conflict graph with the two style information from the candidate style pair as nodes in the multiple style conflict graphs corresponding to the style layer. If so, the style conflict graph is determined as the target style conflict graph. Furthermore, it is determined whether the nodes corresponding to the two style information in the target style conflict graph are connected by an edge.

[0056] Figure 3 This is a schematic diagram of a method for determining a target style conflict map provided in an embodiment of this disclosure, such as... Figure 3 As shown, in some embodiments of this disclosure, determining a target style conflict graph in multiple style conflict graphs, using two style information from candidate style pairs as nodes, includes: Step 301: If there is a first-layer conflict map in the layer conflict map that includes two style information, then the first-layer conflict map is determined as the target style conflict map.

[0057] The first-layer conflict graph can be a layer conflict graph in which the two style information in the candidate style pair are used as two nodes respectively.

[0058] In this embodiment, it is first determined whether there is a first-layer conflict graph in the multiple layer conflict graphs corresponding to multiple style layers, with the two style information in the candidate style pair as nodes. If so, the first-layer conflict graph is determined as the target style conflict graph.

[0059] Step 302: If there is no first-layer conflict graph in the layer conflict graph, then among multiple style layers, the style layer with two style types corresponding to two style information is determined as the target style layer.

[0060] The target style layer can be the style layer to which the two style types corresponding to the two style information belong.

[0061] In this embodiment, if it is determined that a first-layer conflict graph does not exist in any of the multiple layer conflict graphs, or that the style conflict graph does not contain a layer conflict graph, the training device of the click-through rate prediction model can extract two style types from the two style information in the candidate style pair. Among the multiple style layers, the style layer that includes these two style types is determined as the target style layer.

[0062] Step 303: Determine the target style conflict map based on the two style information and at least one set of conflict maps corresponding to the target style layer.

[0063] In this embodiment, the training device for the click-through rate prediction model can index the target style conflict graph in the group conflict graph corresponding to the target style layer based on two style information to determine the target style conflict graph.

[0064] In some embodiments of this disclosure, determining a target style conflict map based on two style information and at least one set of conflict maps corresponding to the target style layer includes: Step a1: If the target style layer is not at the top level, determine the preceding style information in the sample card information that corresponds to the preceding style layer.

[0065] Among them, the preceding style information can be the style information belonging to the preceding style layer in the style information of the sample card information.

[0066] In this embodiment, each non-top-level style layer can correspond to at least one group conflict graph. Each group conflict graph can have a corresponding group conflict graph spectrum, and each group conflict graph corresponds to a preceding style information in the preceding style layer. Therefore, the preceding style information can serve as the index information for the group conflict graph. If the level corresponding to the target style layer is not the top level, the training device of the click-through rate prediction model can determine the preceding style layer located one level above the target style layer, and determine the preceding style information belonging to that preceding style layer in the style information of the sample card information.

[0067] Step a2: In the group conflict graph corresponding to the current style layer, index according to the previous style information to determine the first group conflict graph corresponding to the previous style information.

[0068] The first set of conflict maps can be a set of conflict maps indexed by the preceding style information.

[0069] In this embodiment, for each group of conflict maps in the target style layer, the training device of the click-through rate prediction model can obtain the index information of the group of conflict maps in each preceding style layer. If it is determined that the index information of the group of conflict maps is consistent with the preceding style information, then the group of conflict maps is determined as the first group of conflict maps.

[0070] Step a3: If a second-layer conflict map containing two style information exists in the first set of conflict maps, then the second-layer conflict map is determined as the target style conflict map.

[0071] In this embodiment, the training device of the click-through rate prediction model can determine whether the first set of conflict graphs uses the two style information in the candidate style pair as nodes. If so, the first set of conflict graphs is determined as the target style conflict graph.

[0072] In the above scheme, the target style conflict map in the group conflict map is determined by using the preceding style information as an index.

[0073] Step 203: If there are at least two style information pairs in a candidate style pair that are connected by an edge in the target style conflict graph, then the sample style set is determined to include two style information pairs that have a mutually exclusive relationship.

[0074] In this embodiment, if among the multiple candidate style pairs corresponding to the sample style set, there exists at least one candidate style pair in which two style information are connected by an edge in the target style conflict graph, it indicates that the two style information contained in the candidate style pair have a mutually exclusive relationship and the two style information cannot exist at the same time. Therefore, it is determined that the sample style set includes two style information with a mutually exclusive relationship.

[0075] In the above scheme, by judging whether there are connection nodes corresponding to candidate style pairs in the target style conflict graph, and using candidate style pairs as the judgment unit, a comprehensive screening of candidate card information is achieved.

[0076] In some embodiments of this disclosure, the sample card information further includes a probability adjustment coefficient. Inputting the user characteristics of the first user and multiple sample pattern sets into an initial model, and having the initial model output the predicted click-through rate of the first user, includes: inputting the user characteristics of the first user, multiple sample pattern sets, and probability adjustment coefficients into the initial model, so that the initial model determines an initial click-through rate based on the user characteristics of the first user and multiple sample pattern sets, and multiplying the initial click-through rate by the probability adjustment coefficients to obtain the predicted click-through rate of the first user.

[0077] The probability adjustment coefficient is a factor used to adjust the click probability predicted by the model through a neural network. This coefficient can be manually set or automatically generated, and it can be negatively correlated with the number of times the ad card is displayed; that is, the more times the ad card is displayed, the smaller the probability adjustment coefficient becomes. The initial click-through rate (CTR) is the CTR predicted by the initial model and determined without adjustment by the probability adjustment coefficient.

[0078] In this embodiment, the probability adjustment coefficient can be pre-set in the sample card information. The training device for the click-through rate (CTR) prediction model can extract the probability adjustment coefficient from the sample card information and input the user characteristics of the first user, multiple sample style sets, and the probability adjustment coefficient into the initial model. The initial model predicts the initial CTR based on the user characteristics of the first user and multiple sample style sets, and then multiplies the initial CTR by the probability adjustment coefficient to obtain the predicted CTR. Subsequently, the model parameters are adjusted using a backpropagation algorithm based on the error between the predicted CTR adjusted by the probability adjustment coefficient and the sample CTR until a preset convergence condition is reached, completing the training of the initial model. This trained initial model is then used as the CTR prediction model.

[0079] In the above scheme, by increasing the probability adjustment coefficient, the risk of overfitting of the model to make high-frequency recommendations of the same ad cards is reduced, and the diversity of ad card display is improved.

[0080] The training scheme for the click-through rate (CTR) prediction model provided in this embodiment obtains sample card information of sample ad cards, user characteristics of a first user, sample CTR of the first user on the sample ad cards, and multiple style conflict graphs corresponding to multiple style layers. The sample card information includes multiple style information, the multiple style layers are functional hierarchies of the style information, and the style conflict graphs are used to record the mutual exclusion relationships between style information. Information extraction is performed on the sample card information according to the multiple style layers to obtain multiple sample style sets corresponding to the multiple style layers. Each sample style set includes style information corresponding to one style layer. If all sample style sets pass the detection of the corresponding style conflict graphs, the user characteristics of the first user and the multiple sample style sets are input into the initial model, and the initial model outputs the predicted CTR of the first user. The model parameters of the initial model are adjusted according to the predicted CTR of the first user and the sample CTR until a preset convergence condition is reached, at which point the initial model is used as the CTR prediction model.

[0081] Using the above technical solution, multiple style layers are obtained according to function, and the mutual exclusion relationship between style information within each style layer is recorded through a style conflict graph. Sample card information is extracted according to style layers to obtain the sample style set corresponding to each style layer. The sample style set of the corresponding style layer is detected by the style conflict graph, avoiding consideration of the mutual exclusion relationship between style information of different style layers, realizing hierarchical isolation of style conflicts, and reducing the number of style conflict combinations to be considered. The initial model is trained based on the multiple sample style sets that have passed the detection, the user characteristics of the first user, and the sample click-through rate, to obtain a model that can predict the click-through rate based on the sample style sets that have passed the detection and user characteristics. The trained model can predict a reasonable click-through rate for ad cards that do not have mutual exclusion style information within the style layer based on user characteristics, and predict a lower click-through rate for ad cards that have mutual exclusion style information. This avoids the ad cards displayed to users containing styles with poor matching effects, and achieves the goal of ensuring the matching effect between different styles in the ad cards while displaying ad cards that match the user characteristics in real time.

[0082] In some embodiments of this disclosure, the training method for the click-through rate prediction model further includes: acquiring multiple display anomaly logs corresponding to multiple historical ad cards; extracting and processing the multiple display anomaly logs to obtain multiple style information pairs; identifying style information pairs whose statistical count exceeds a preset threshold as anomaly information pairs; and incorporating the anomaly information pairs into a style conflict graph for representation to obtain an updated style conflict graph.

[0083] Among them, historical ad cards can be ad cards that were displayed to users at historical moments and exhibited display anomalies. Display anomaly logs record historical ad card display anomalies, such as an ad card crashing unexpectedly after being triggered. Style information pairs can be sets containing two style information pieces. Anomaly information pairs can be sets containing two mutually exclusive style information pieces, determined based on statistical data. The statistical count is the number of times a style information pair appears. The preset quantity threshold is a pre-set threshold for the statistical count.

[0084] In this embodiment, the training device for the click-through rate prediction model can acquire historical ad cards and their corresponding display anomaly logs, and extract style information pairs from the display anomaly logs using a pre-set information pair extraction algorithm. This embodiment does not limit the information pair extraction algorithm; for example, it can be an algorithm based on a natural language model. Further, the occurrence frequency of each style information pair is counted to obtain the statistical quantity of each style information pair. It is determined whether the statistical quantity corresponding to each style information pair is greater than a preset quantity threshold; if so, the style information pair is identified as an anomaly information pair. If the two style types corresponding to the two style information pieces in the anomaly information pair correspond to the same style layer, the two style information pieces in the anomaly information pair are used as two nodes in a sub-graph to generate a sub-graph. Further, the sub-graph is fused with the style conflict graph of the corresponding style layer to obtain an updated style conflict graph.

[0085] The above scheme enables the automatic exploration of potential conflicting style information pairs and the dynamic updating of the style conflict map, thereby reducing the manual maintenance cost of the style conflict map.

[0086] Figure 4 This is a flowchart illustrating a method for displaying an advertising card according to an embodiment of this disclosure. The method can be executed by an advertising card display device, which can be implemented using software and / or hardware, and is generally integrated into an electronic device. This method can be applied to electronic devices deployed with click-through rate prediction models. This electronic device can be a server-side electronic device. Figure 4 As shown, the method includes: Step 401: In response to the ad display request sent by the client, obtain the user characteristics of the target user and the information of the ad card to be analyzed.

[0087] The ad display request can be a request to display an ad card, generated in response to a user's ad display action. This action may include page loading, pre-loading preparations, etc. The target user can be the user currently displaying the ad card. The target user's characteristics are obtained with their authorization. The ad card to be analyzed can be the ad card for which click-through rate prediction is to be performed. The card information to be analyzed can be the card information of the ad card to be analyzed.

[0088] In this embodiment of the disclosure, a user can perform a card display operation on the client. The client, in response to the advertising card display operation, generates an advertising display request and sends the request to the advertising card display device. The advertising card display device, in response to the advertising display request, obtains the target user's user characteristics and the card information to be analyzed for the advertising card to be analyzed.

[0089] Step 402: Extract information from the card information to be analyzed by layer according to multiple style layers to obtain multiple sets of styles to be analyzed corresponding to multiple style layers.

[0090] A set of styles to be analyzed can include the style information corresponding to the card information to be analyzed in a style layer.

[0091] In this embodiment, for each style layer, the display device of the advertising card can extract the style information belonging to that style layer from the style information included in the card information to be analyzed, and obtain the set of styles to be analyzed corresponding to that style layer.

[0092] Step 403: Input the user characteristics of the target user and multiple sets of styles to be analyzed into the click-through rate prediction model, and the click-through rate prediction model outputs the predicted click-through rate of the target user; the click-through rate prediction model is trained by the click-through rate prediction model training method.

[0093] In this embodiment, multiple sets of styles to be analyzed are characterized using methods such as self-attention encoding to obtain corresponding features, which can be normalized vectors. Furthermore, the user features of the target user and the features corresponding to the multiple sets of styles to be analyzed are input into the click-through rate (CTR) prediction model, and the predicted CTR is calculated through the CTR prediction model.

[0094] Step 404: Determine the target card information in the card information to be analyzed based on the predicted click-through rate.

[0095] The target card information can be used to generate advertising cards that are displayed to users.

[0096] In this embodiment, the advertising card display device can filter the card information to be analyzed based on the predicted click-through rate to obtain target card information. This embodiment does not limit the method of determining target card information from the card information to be analyzed based on the predicted click-through rate. For example, after sorting the card information to be analyzed in descending order of predicted click-through rate, the card information ranked before a preset order can be determined as candidate card information. Alternatively, card information with a preset click-through rate greater than a click-through rate threshold can be determined as target card information.

[0097] Step 405: Send the target card information to the client so that the target advertising card is displayed on the client's page based on the target card information.

[0098] Among them, the target ad card can be the ad card that is rendered and displayed to the user.

[0099] In this embodiment of the disclosure, after determining the target card information, if the number of target card information is one, the advertising card display device can send the target card information to the client. After receiving the target card information, the client renders the target card information on the page and then displays the target advertising card on the page. If the number of target information cards is multiple, the advertising card display device can select one target information card from the target advertising cards according to the predicted click probability, send the one target card information to the client, and after receiving the target card information, the client renders the one target card information on the page and then displays the one target advertising card on the page.

[0100] The advertising card display method provided in this disclosure is applied to an electronic device equipped with a click-through rate (CTR) prediction model. The method includes: responding to an advertising display request sent by a client; obtaining user characteristics of a target user and information about the advertising card to be analyzed; performing layered information extraction on the information about the advertising card to be analyzed according to multiple style layers to obtain multiple sets of styles to be analyzed corresponding to the multiple style layers; inputting the user characteristics of the target user and the multiple sets of styles to be analyzed into the CTR prediction model, and having the CTR prediction model output the predicted CTR of the target user; the CTR prediction model is trained using a CTR prediction model training method; determining the target card information in the information about the advertising card to be analyzed based on the predicted CTR; and sending the target card information to the client so that the target advertising card is displayed on the client's page according to the target card information.

[0101] Using the above technical solution, the information of the card to be analyzed is extracted according to the style layer, resulting in multiple sets of styles to be analyzed corresponding to multiple style layers. These multiple sets of styles to be analyzed, along with the user characteristics of the target user, are input into a click-through rate (CTR) prediction model. This CTR prediction model can predict a reasonable CTR for ad cards that do not have mutually exclusive style information within the style layer, and a lower CTR for ad cards that do have mutually exclusive style information, based on the user characteristics. The target card information in the card information to be analyzed is determined based on the predicted CTR, and the target ad card is displayed on the client side based on this target card information. This avoids displaying ad cards with poor matching effects to the user, ensuring the matching effect between different styles in the ad card while displaying ad cards that match the user characteristics in real time.

[0102] Figure 5 A flowchart illustrating another method for displaying an advertising card provided in this disclosure embodiment is shown below. Figure 5 As shown, the method for displaying the advertising card before sending the target card information to the client also includes: Step 501: Obtain a set of multiple target styles corresponding to the target card information.

[0103] A target style set can include the style information corresponding to the target card information in a style layer.

[0104] In this embodiment of the disclosure, for each style layer, the display device of the advertising card can extract the style information belonging to that style layer from the style information included in the target card information to obtain the target style set corresponding to that style layer.

[0105] Step 502: For each target style set, if the target style set is determined to include two mutually exclusive style information based on the corresponding style conflict map, then the target style set is determined to have failed the detection; otherwise, the target style set is determined to have passed the detection.

[0106] In this embodiment, for each target style set, the advertising card display device can extract the style information within the target style set, match the extracted style information with the style conflict graph corresponding to the style layer, and if it is determined that there are two style information as two nodes connected by an edge in the style conflict graph, then it is determined that the target style set has failed the detection; otherwise, it is determined that the target style set has passed the detection.

[0107] Step 503: If at least one target style set fails the detection, delete the target card information.

[0108] In this embodiment, if it is determined that at least one target style set among the multiple target style sets corresponding to the target card information fails the detection, it indicates that there are mutually exclusive style information in the target card information, and the target card information is then deleted.

[0109] In the above solution, the target card information is checked a second time through the style conflict graph, which ensures that the target ad cards displayed to each user do not contain style information with mutual exclusion relationships recorded in the style conflict graph, thereby improving the user's viewing experience.

[0110] The following example will further illustrate the method for displaying advertising cards in this embodiment of the disclosure. Figure 6 A schematic diagram illustrating a method for displaying an advertising card according to an embodiment of this disclosure, as shown below. Figure 6 As shown, the methods for displaying this advertising card include: First, the style layers are determined, and a style conflict graph corresponding to each style layer is constructed. Specifically, style information is determined, which serves as the smallest unit of interaction with the user during the recommendation display process. Pre-defined style layers are obtained, and the style information is categorized according to the style types included in each style layer. Sub-graphs are generated based on the mutual exclusion rules between style information, or by mining implicit conflict information from display anomaly logs. Graph fusion processing is then performed on the sub-graphs corresponding to the same style layer to generate the style conflict graph corresponding to that style layer.

[0111] Further, sample card information is generated, and the initial model is trained based on this sample card information to obtain a click-through rate (CTR) prediction model. Specifically, style information is traversed and combined to obtain multiple sample card information sets, and the sample card information is filtered using a style conflict graph to remove sample card information containing conflicting style information. Sample user features and sample card information are combined one by one, and a score is assigned to each set of sample user features and sample card information to determine the sample CTR, or the sample CTR is determined based on historical CTR. The sample conflict graph is injected as prior knowledge into message passing, and self-attention encoding is applied to the sample card information. A weighted cross-entropy loss function is used to train the initial model, resulting in a fully trained CTR prediction model.

[0112] Furthermore, target user characteristics are acquired, and target card information is determined from the card information to be analyzed using a click-through rate prediction model. The target ad card is then rendered on the page based on this target card information. Specifically, the current user is taken as the target user, the click-through rate prediction model is invoked to determine the target card information from the card information to be analyzed, and a style conflict graph is used to perform secondary validation on the target card information. Target card information with conflicting relationships is removed, and the target ad card is rendered and displayed on the page based on the retained target card information.

[0113] Furthermore, incremental training data is determined based on the target ad card, and the click-through rate (CTR) prediction model is trained using this incremental training data. Specifically, after completing the interaction with the target ad card, the target card information, target user characteristics, and the target CTR determined based on the interaction are identified as incremental training data. This incremental training data is then used to train the CTR prediction model, enabling online learning and continuous iteration of the CTR prediction model.

[0114] This embodiment of the disclosure adopts a modular structure comprising a base layer, a content layer, and an interaction layer. It accurately models the mutual exclusion relationships between style information using a style conflict graph, fundamentally resolving the style information conflict problem. A neural network integrating multi-source data annotation and conflict perception enables personalized matching of ad style information with user characteristics, significantly improving click-through rates and user experience. Furthermore, it can be widely applied to different advertising scenarios, offering significant advantages such as strong scalability, rapid adaptation to diverse user needs, and consistent and aesthetically pleasing display effects.

[0115] Figure 7 This is a schematic diagram of a training device for a click-through rate prediction model provided in an embodiment of this disclosure. This device can be implemented using software and / or hardware. Figure 7 As shown, the training apparatus for this click-through rate prediction model includes: The first acquisition module 701 is used to acquire sample card information of sample advertising cards, user characteristics of the first user, sample click-through rate of the first user on the sample advertising cards, and multiple style conflict graphs corresponding to multiple style layers; wherein, the sample card information includes multiple style information, the multiple style layers are functional layers of the style information, and the style conflict graph is used to record the mutual exclusion relationship between style information. The first extraction module 702 is used to extract information from the sample card information in layers according to the multiple style layers, so as to obtain multiple sample style sets corresponding to the multiple style layers; wherein, each sample style set includes style information corresponding to one style layer; The training module 703 is used to input the user features of the first user and the multiple sample style sets into an initial model if all of the multiple sample style sets pass the detection of the corresponding style conflict map, and output the predicted click-through rate of the first user from the initial model. The model parameters of the initial model are adjusted according to the predicted click-through rate of the first user and the sample click-through rate until a preset convergence condition is reached, and the initial model is used as the click-through rate prediction model.

[0116] Optionally, the pattern conflict map includes a layer conflict map and / or a group conflict map; The layer conflict graph corresponds one-to-one with the style layer, and the layer conflict graph is used to record the common mutual exclusion relationship between style information within the corresponding style layer; The group conflict graph corresponds to the information group in the style layer. Each information group includes multiple style information in the corresponding style layer. Each information group corresponds to specific style information in the preceding style layer of the style layer. The preceding style layer is the style layer at the next higher level. The group conflict graph is used to record the mutual exclusion relationship between style information within the information group.

[0117] Optionally, the step of extracting information by layering the sample card information according to the multiple style layers to obtain multiple sample style sets corresponding to the multiple style layers includes: For each style layer, among the multiple style information corresponding to the sample card information, the style information whose style type matches the style layer is assigned to the sample style set corresponding to the style layer.

[0118] Optionally, all of the multiple sample style sets are detected through corresponding style conflict maps, including: For each of the sample style sets, if the sample style set is determined to include two style information with a mutual exclusion relationship according to the corresponding style conflict map, then the sample style set is determined to have failed the detection; otherwise, the sample style set is determined to have passed the detection.

[0119] Optionally, determining that the sample style set includes two mutually exclusive style information items based on the corresponding style conflict map includes: Extract pairwise style information from the sample style set to form multiple candidate style pairs; For each candidate style pair, determine the target style conflict graph in the plurality of style conflict graphs that uses the two style information in the candidate style pair as nodes, and determine whether the two style information in the target style conflict graph are connected by an edge. If there exists at least one candidate style pair in which two style information are connected by an edge in the target style conflict graph, then the sample style set is determined to include two style information with a mutually exclusive relationship.

[0120] Optionally, determining the target style conflict graph in the plurality of style conflict graphs, which uses two style information from the candidate style pair as nodes, includes: If a first-layer conflict map containing the two style information exists in the layer conflict map, then the first-layer conflict map is determined as the target style conflict map; If the first layer conflict graph does not exist in the layer conflict graph, then among multiple style layers, the style layer corresponding to the two style types of the two style information is determined as the target style layer; The target style conflict map is determined based on the two style information and at least one set of conflict maps corresponding to the target style layer.

[0121] Optionally, determining the target style conflict map based on the two style information and at least one set of conflict maps corresponding to the target style layer includes: If the target style layer is not at the top level, determine the preceding style information in the sample card information that corresponds to the preceding style layer. In the group conflict graph corresponding to the current style layer, the first group conflict graph corresponding to the preceding style information is determined by indexing according to the preceding style information. If a second-layer conflict map containing the two style information exists in the first set of conflict maps, then the second-layer conflict map is determined as the target style conflict map.

[0122] Optionally, the sample card information further includes a probability adjustment coefficient, and the step of inputting the user characteristics of the first user and the multiple sample pattern sets into the initial model, and having the initial model output the predicted click-through rate of the first user, includes: The user characteristics of the first user, the multiple sample pattern sets, and the probability adjustment coefficient are input into the initial model so that the initial model determines the initial click-through rate based on the user characteristics of the first user and the multiple sample pattern sets, and multiplies the initial click-through rate and the probability adjustment coefficient to obtain the predicted click-through rate of the first user.

[0123] Optionally, the training apparatus for the click-through rate prediction model further includes an update module, which is used to: Retrieve multiple display anomaly logs corresponding to multiple historical ad cards; The multiple display error logs are extracted and processed to obtain multiple style information pairs; The style information pairs whose statistical count exceeds a preset threshold are identified as abnormal information pairs. The abnormal information is incorporated into the style conflict graph to obtain the updated style conflict graph.

[0124] The training apparatus for the click-through rate prediction model provided in this disclosure can execute the training method for the click-through rate prediction model provided in any embodiment of this disclosure, and has the corresponding functional modules and beneficial effects of the execution method.

[0125] Figure 8This is a schematic diagram of the structure of an advertising card display device provided in an embodiment of the present disclosure. The device can be implemented by software and / or hardware. For example... Figure 8 As shown, the display device for the advertising card, located on an electronic device equipped with a click-through rate prediction model, includes: The second acquisition module 801 is used to respond to the advertisement display request sent by the client and acquire the user characteristics of the target user and the information of the advertisement card to be analyzed. The second extraction module 802 is used to perform layered information extraction on the card information to be analyzed according to multiple style layers, so as to obtain multiple sets of styles to be analyzed corresponding to the multiple style layers. The prediction module 803 is used to input the user characteristics of the target user and the multiple sets of styles to be analyzed into the click-through rate prediction model, and the click-through rate prediction model outputs the predicted click-through rate of the target user; the click-through rate prediction model is trained by the click-through rate prediction model training method. The determination module 804 is used to determine the target card information in the card information to be analyzed based on the predicted click-through rate; The display module 805 is used to send the target card information to the client so that the target advertising card is displayed on the client's page according to the target card information.

[0126] The advertising card display device provided in this disclosure can execute the advertising card display method provided in any embodiment of this disclosure, and has the corresponding functional modules and beneficial effects of the execution method.

[0127] This disclosure also provides a computer program product, including a computer program / instructions, which, when executed by a processor, implement the training method for the click-through rate prediction model and / or the display method for advertising cards provided in any embodiment of this disclosure.

[0128] Figure 9 This is a schematic diagram of an electronic device provided in an embodiment of the present disclosure. See below for details. Figure 9 The diagram illustrates a structural schematic suitable for implementing the electronic device 900 in the embodiments of this disclosure. The electronic device 900 in the embodiments of this disclosure may include, but is not limited to, mobile terminals such as mobile phones, laptops, digital broadcast receivers, PDAs (personal digital assistants), PADs (tablet computers), PMPs (portable multimedia players), in-vehicle terminals (e.g., in-vehicle navigation terminals), and fixed terminals such as digital TVs and desktop computers. Figure 9 The electronic device shown is merely an example and should not be construed as limiting the functionality and scope of the embodiments disclosed herein.

[0129] like Figure 9As shown, electronic device 900 may include a processing device (e.g., a central processing unit, a graphics processor, etc.) 901, which can perform various appropriate actions and processes according to a program stored in read-only memory (ROM) 902 or a program loaded from storage device 908 into random access memory (RAM) 903. RAM 903 also stores various programs and data required for the operation of electronic device 900. Processing device 901, ROM 902, and RAM 903 are interconnected via bus 904. Input / output (I / O) interface 905 is also connected to bus 904.

[0130] Typically, the following devices can be connected to I / O interface 905: input devices 906 including, for example, touchscreens, touchpads, keyboards, mice, cameras, microphones, accelerometers, gyroscopes, etc.; output devices 907 including, for example, liquid crystal displays (LCDs), speakers, vibrators, etc.; storage devices 908 including, for example, magnetic tapes, hard disks, etc.; and communication devices 909. Communication device 909 allows electronic device 900 to communicate wirelessly or wiredly with other devices to exchange data. Although Figure 9 An electronic device 900 with various devices is shown; however, it should be understood that it is not required to implement or possess all of the devices shown. More or fewer devices may be implemented or possessed alternatively.

[0131] In particular, according to embodiments of this disclosure, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of this disclosure include a computer program product comprising a computer program carried on a non-transitory computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication device 909, or installed from a storage device 908, or installed from a ROM 902. When the computer program is executed by the processing device 901, it performs the functions defined in the click-through rate prediction model training method and / or the ad card display method of embodiments of this disclosure.

[0132] It should be noted that the computer-readable medium described in this disclosure can be a computer-readable signal medium or a computer-readable storage medium, or any combination thereof. A computer-readable storage medium can be, for example,—but not limited to—an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of a computer-readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In this disclosure, a computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. In this disclosure, a computer-readable signal medium can include a data signal propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals can take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A computer-readable signal medium can be any computer-readable medium other than a computer-readable storage medium, which can send, propagate, or transmit a program for use by or in connection with an instruction execution system, apparatus, or device. The program code contained on the computer-readable medium can be transmitted using any suitable medium, including but not limited to: wires, optical fibers, RF (radio frequency), etc., or any suitable combination thereof.

[0133] In some implementations, clients and servers can communicate using any currently known or future-developed network protocol such as HTTP (Hypertext Transfer Protocol) and can interconnect with digital data communication (e.g., communication networks) of any form or medium. Examples of communication networks include local area networks (“LANs”), wide area networks (“WANs”), the Internet (e.g., the Internet of Things), and peer-to-peer networks (e.g., ad hoc peer-to-peer networks), as well as any currently known or future-developed networks.

[0134] The aforementioned computer-readable medium may be included in the aforementioned electronic device; or it may exist independently and not assembled into the electronic device.

[0135] The aforementioned computer-readable medium carries one or more programs, which, when executed by the electronic device, cause the electronic device to implement the click-through rate prediction model training method and / or the advertising card display method of the embodiments of this disclosure.

[0136] Computer program code for performing the operations of this disclosure can be written in one or more programming languages ​​or a combination thereof, including but not limited to object-oriented programming languages ​​such as Java, Smalltalk, and C++, as well as conventional procedural programming languages ​​such as the "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0137] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this disclosure. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0138] The units described in the embodiments of this disclosure can be implemented in software or hardware. The names of the units are not, in some cases, intended to limit the specific unit.

[0139] The functions described above in this document can be performed at least in part by one or more hardware logic components. For example, exemplary types of hardware logic components that can be used, without limitation, include: field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), system-on-a-chip (SoCs), complex programmable logic devices (CPLDs), and so on.

[0140] In the context of this disclosure, a machine-readable medium can be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium can be, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0141] It is understood that before using the technical solutions disclosed in the various embodiments of this disclosure, users should be informed of the types, scope of use, and usage scenarios of the information involved in this disclosure in an appropriate manner in accordance with relevant laws and regulations, and user authorization should be obtained.

[0142] The above description is merely a preferred embodiment of this disclosure and an explanation of the technical principles employed. Those skilled in the art should understand that the scope of this disclosure is not limited to technical solutions formed by specific combinations of the above-described technical features, but should also cover other technical solutions formed by arbitrary combinations of the above-described technical features or their equivalents without departing from the above-described concept. For example, technical solutions formed by substituting the above features with (but not limited to) technical features disclosed in this disclosure that have similar functions.

[0143] Furthermore, while the operations are described in a specific order, this should not be construed as requiring these operations to be performed in the specific order shown or in a sequential order. In certain environments, multitasking and parallel processing may be advantageous. Similarly, while several specific implementation details are included in the above discussion, these should not be construed as limiting the scope of this disclosure. Certain features described in the context of individual embodiments may also be implemented in combination in a single embodiment. Conversely, various features described in the context of a single embodiment may also be implemented individually or in any suitable sub-combination in multiple embodiments.

[0144] Although the subject matter has been described using language specific to structural features and / or methodological logic, it should be understood that the subject matter defined in the appended claims is not necessarily limited to the specific features or actions described above. Rather, the specific features and actions described above are merely illustrative examples of implementing the claims.

Claims

1. A training method for a click-through rate prediction model, characterized in that, include: The sample card information of the sample ad card, the user characteristics of the first user, the sample click-through rate of the first user on the sample ad card, and multiple style conflict graphs corresponding to multiple style layers are obtained; wherein, the sample card information includes multiple style information, the multiple style layers are functional layers of the style information, and the style conflict graph is used to record the mutual exclusion relationship between style information. Based on the multiple style layers, the sample card information is extracted hierarchically to obtain multiple sample style sets corresponding to the multiple style layers; wherein, each sample style set includes style information corresponding to one style layer; If all of the multiple sample style sets pass the detection of the corresponding style conflict graph, then the user features of the first user and the multiple sample style sets are input into the initial model, and the initial model outputs the predicted click-through rate of the first user. The model parameters of the initial model are adjusted according to the predicted click-through rate of the first user and the sample click-through rate until a preset convergence condition is reached, and then the initial model is used as the click-through rate prediction model.

2. The method of claim 1, wherein, The pattern conflict map includes layer conflict maps and / or group conflict maps; The layer conflict graph corresponds one-to-one with the style layer, and the layer conflict graph is used to record the common mutual exclusion relationship between style information within the corresponding style layer; The group conflict graph corresponds to the information group in the style layer. Each information group includes multiple style information in the corresponding style layer. Each information group corresponds to specific style information in the preceding style layer of the style layer. The preceding style layer is the style layer at the next higher level. The group conflict graph is used to record the mutual exclusion relationship between style information within the information group.

3. The method of claim 1, wherein, The step of extracting information from the sample card information in layers according to the multiple style layers to obtain multiple sample style sets corresponding to the multiple style layers includes: For each style layer, among the multiple style information corresponding to the sample card information, the style information whose style type matches the style layer is assigned to the sample style set corresponding to the style layer.

4. The method of claim 1, wherein, The multiple sample style sets were all detected through the corresponding style conflict maps, including: For each of the sample style sets, if the sample style set is determined to include two style information with a mutual exclusion relationship according to the corresponding style conflict map, then the sample style set is determined to have failed the detection; otherwise, the sample style set is determined to have passed the detection.

5. The method of claim 3, wherein, The step of determining that the sample style set includes two mutually exclusive style information based on the corresponding style conflict map includes: Extract pairwise style information from the sample style set to form multiple candidate style pairs; For each candidate style pair, determine the target style conflict graph in the plurality of style conflict graphs that uses the two style information in the candidate style pair as nodes, and determine whether the two style information in the target style conflict graph are connected by an edge. If there exists at least one candidate style pair in which two style information are connected by an edge in the target style conflict graph, then the sample style set is determined to include two style information with a mutually exclusive relationship.

6. The method of claim 5, wherein, Determining the target style conflict graph in the plurality of style conflict graphs, which uses two style information from the candidate style pair as nodes, includes: If a first-layer conflict map containing the two style information exists in the layer conflict map, then the first-layer conflict map is determined as the target style conflict map; If the first layer conflict graph does not exist in the layer conflict graph, then among multiple style layers, the style layer corresponding to the two style types of the two style information is determined as the target style layer; The target style conflict map is determined based on the two style information and at least one set of conflict maps corresponding to the target style layer.

7. The method according to claim 6, characterized in that, The step of determining the target style conflict map based on the two style information and at least one set of conflict maps corresponding to the target style layer includes: If the target style layer is not at the top level, determine the preceding style information in the sample card information that corresponds to the preceding style layer. In the group conflict graph corresponding to the current style layer, the first group conflict graph corresponding to the preceding style information is determined by indexing according to the preceding style information. If a second-layer conflict map containing the two style information exists in the first set of conflict maps, then the second-layer conflict map is determined as the target style conflict map.

8. The method according to claim 1, characterized in that, The sample card information also includes a probability adjustment coefficient. The step of inputting the user characteristics of the first user and the multiple sample pattern sets into the initial model, and having the initial model output the predicted click-through rate of the first user, includes: The user characteristics of the first user, the multiple sample pattern sets, and the probability adjustment coefficient are input into the initial model so that the initial model determines the initial click-through rate based on the user characteristics of the first user and the multiple sample pattern sets, and multiplies the initial click-through rate and the probability adjustment coefficient to obtain the predicted click-through rate of the first user.

9. The method of claim 1, wherein, The method further includes: Retrieve multiple display anomaly logs corresponding to multiple historical ad cards; The multiple display error logs are extracted and processed to obtain multiple style information pairs; The style information pairs whose statistical count exceeds a preset threshold are identified as abnormal information pairs. The abnormal information is incorporated into the style conflict graph to obtain the updated style conflict graph.

10. A method of displaying advertising cards, characterized by The method, applied to an electronic device equipped with a click-through rate prediction model, includes: In response to an ad display request sent by the client, obtain the user characteristics of the target user and the information of the ad card to be analyzed; Based on the layered information extraction of the card information to be analyzed according to multiple style layers, multiple sets of styles to be analyzed corresponding to the multiple style layers are obtained; The user characteristics of the target user and the multiple sets of styles to be analyzed are input into the click-through rate prediction model, and the click-through rate prediction model outputs the predicted click-through rate of the target user; the click-through rate prediction model is trained by the click-through rate prediction model training method of any one of claims 1-9; The target card information in the card information to be analyzed is determined based on the predicted click-through rate. The target card information is sent to the client so that the target advertising card is displayed on the client's page based on the target card information.

11. The method according to claim 10, characterized in that, Before sending the target card information to the client, the method further includes: Obtain multiple target style sets corresponding to the target card information; For each target style set, if the target style set is determined to include two mutually exclusive style information based on the corresponding style conflict map, then the target style set is determined to have failed the detection; otherwise, the target style set is determined to have passed the detection. If at least one of the target style sets fails the detection, the target card information is deleted.

12. A training device for a click-through rate prediction model, characterized in that, The device includes: The first acquisition module is used to acquire sample card information of sample advertising cards, user characteristics of the first user, sample click-through rate of the first user on the sample advertising cards, and multiple style conflict graphs corresponding to multiple style layers; wherein, the sample card information includes multiple style information, the multiple style layers are functional layers of the style information, and the style conflict graph is used to record the mutual exclusion relationship between style information. The first extraction module is used to extract information from the sample card information in layers according to the multiple style layers, so as to obtain multiple sample style sets corresponding to the multiple style layers; wherein, each sample style set includes style information corresponding to one style layer; The training module is used to input the user features of the first user and the multiple sample style sets into an initial model if all of the multiple sample style sets pass the detection of the corresponding style conflict map. The initial model outputs the predicted click-through rate of the first user. The model parameters of the initial model are adjusted according to the predicted click-through rate of the first user and the sample click-through rate until a preset convergence condition is reached. The initial model is then used as the click-through rate prediction model.

13. An advertising card display device, characterized by comprising: The device is configured on an electronic device equipped with a click-through rate prediction model, the device comprising: The second acquisition module is used to respond to the ad display request sent by the client and acquire the user characteristics of the target user and the information of the ad card to be analyzed. The second extraction module is used to extract information from the card information to be analyzed in layers according to multiple style layers, so as to obtain multiple sets of styles to be analyzed corresponding to the multiple style layers. The prediction module is used to input the user characteristics of the target user and the multiple sets of styles to be analyzed into the click-through rate prediction model, and the click-through rate prediction model outputs the predicted click-through rate of the target user; the click-through rate prediction model is trained by the click-through rate prediction model training method of any one of claims 1-9; The determination module is used to determine the target card information in the card information to be analyzed based on the predicted click-through rate; The display module is used to send the target card information to the client so that the target advertising card is displayed on the client's page according to the target card information.

14. An electronic device, comprising: The electronic device includes: processor; Memory used to store the processor's executable instructions; The processor is configured to read the executable instructions from the memory and execute the instructions to implement the training method of the click-through rate prediction model as described in any one of claims 1-9 or the display method of the advertising card as described in any one of claims 10-11.

15. A computer-readable storage medium, characterized in that, The storage medium stores a computer program for executing the training method of the click-through rate prediction model as described in any one of claims 1-9 or the display method of the advertising card as described in any one of claims 10-11.

Images